Eugen Leitl wrote: > On Sat, 30 Nov 2002, Morlock Elloi wrote: > >>> 1. large wifi networks start to hit scaling problems - they start >>> to need routers and name services that are relatively expensive, >>> and ip address > Geographic routing completely eliminates need for expensive routing > and admin traffic. Name services? Who needs name services? Localhost > is sufficient for a prefix to an address namepace. without routing and name services, you have what amounts to a propriatory NAT solution - no way to address an interior node on the cloud from the internet (and hence, peer to peer services or any other protocol that requires an inbound connection not directly understood by the nat translation - eg ftp on a non standard port or ssl-encrypted as ftps)
> Actually, even a MAC has enough address space to label entire Earth > surface with ~1 address/m^2, IPv6 addresses are plenty better here. > And of course no one forces you to use actual IP addresses. You can > sure tunnel TCP/IP through a geographic routing protocol. under ipv6 you can avoid having to have a explicit naming service - the cloud id of the card (possibly with a network prefix to identify the cloud as a whole) can *be* the unique name; routing is still an issue but that reduces to being able to route to a unique node inside the cloud - which appears from a brief glance at the notes from Morlock Elloi (thanks again :) to have at least a workable trial solution. if a IPv6 internet ever becomes a reality, clouds would fit right in. TCP/IP tunnelling without a name service at at least one end isn't workable; *static* NAT/PAT is of course a name service and can't be considered, but SOCKS and socks aware p2p is a definite possibility.
